Xiaomi's MiMo team revealed training progress on MiMo-v2.6 models (pro and flash variants), showing DeepSWE benchmark scores exceeding 60%, on par with GLM-5.3-Flash. The v2.5 model remains popular on OpenRouter (ranked 6th) due to cost advantages, particularly low cache pricing. Latest training dashboard shows reinforcement learning costs over $1.2M, though full model training likely costs substantially more. Significant improvement trajectory suggests competitive positioning in the LLM market.